利用全景图的视频稳像算法

针对具有场景变换、景深变换较小的抖动视频，本文提出了一种利用全景图的视频稳像方法。不同于传统稳像方法，本文所提算法利用原始视频序列生成广角视图，通过后期抽取视频帧，以合成方式输出稳定视频。通过区域块的运动估计，获取相机的运动轨迹；在此基础上，为消除视频抖动导致的原始运动轨迹的非平滑问题，提出了一种基于运动矢量统计的相机运动模式判决方法，并以此选择相应的平滑方式。最后，通过平滑后的运动轨迹对全景图进行抽帧，合成平滑、稳定视频。实验结果表明，本文提出的稳像方法能有效去除较大抖动，同时避免了对视频宽度的裁剪等后处理，具有较好的稳像能力。 

Video Stabilization Algorithm Using Generated Panorama

To stabilize shaky video sequences with scene change and less depth variation, we develop a video stabilization approach using generated panorama. Different from traditional stabilization techniques, we took advantage of original input videos to generate panorama and then captured frames to synthetize the output stabilized video. To obtain the camera motion trajectory, block matching based motion estimation method was first adopted. On this basis, we proposed a judging criterion based on motion vector statistics to remove the jitter of original motion trajectory. Thus, with the determined motion pattern of camera, a specific smoothing method can be applied. Given the smoothed motion trajectory, a stabilized video can be generated by subsampling on the synthesized panorama image. Experimental results have illustrated that the proposed approach can well deal with shaky videos with large jitter. Especially, without further using post-processing like clipping, the generated video looks quite natural.